Independent tuning of size and coverage of supported Pt nanoparticles using atomic layer deposition

Synthetic methods that allow for the controlled design of well-defined Pt nanoparticles are highly desirable for fundamental catalysis research. In this work, we propose a strategy that allows precise and independent control of the Pt particle size and coverage. Our approach exploits the versatility of the atomic layer deposition (ALD) technique by combining two ALD processes for Pt using different reactants. The particle areal density is controlled by tailoring the number of ALD cycles using trimethyl(methylcyclopentadienyl)platinum and oxygen, while subsequent growth using the same Pt precursor in combination with nitrogen plasma allows for tuning of the particle size at the atomic level. The excellent control over the particle morphology is clearly demonstrated by means of in situ and ex situ X-ray fluorescence and grazing incidence small angle X-ray scattering experiments, providing information about the Pt loading, average particle dimensions, and mean center-to-center particle distance.

The manuscript reports on that what is exactly expressed by the title: The independent tuning of size and coverage of supported Pt nanoparticles using ALD. In principle this is an interesting feature that builds on extensive other work that has been published in this area: -The preparation of supported nanoparticles by ALD for catalysis applications -The precise size-control of these nanoparticles (also for particles smaller than reported here) -The preparation of nanoparticles consisting of several materials (Pt, Ru, Pd), also in mixed phases (alloys) or in core/shell configurations -The preparation of nanoparticles on highly structured materials (e.g. on nanosphere supports) -The protection of the nanoparticles by overcoatings -The area-selective ALD of nanoparticles -The demonstration of the activity of ALD-prepared nanoparticles in several heterogenous catalysis reactions (viz. various dehydrogenation and oxidation reactions) -Etc. This work goes back to the basis of the field and the (only) novelty of the work is that the authors show that the use of a N2* plasma as reactant allows for increasing the size of the Pt nanoparticles without changing the coverage of the Pt nanoparticles in terms of the number of particles per surface area. With the common chemistry using O2 as the reactant, the size can be controlled but not without affecting the coverage of the nanoparticles.
As mentioned, this is a nice feature but by itself it does not warrant the publication of the work in Nature Communications. More important achievements (see above) have already been reported in high impact journals (including several Nature and Science journals). Furthermore, it is not clear what the impact of this work exactly is. The size of the Pt nanoparticles synthesized is relatively large (for optimized catalytic reactivity) whereas the 2-step method does not allow to increase the coverage of the nanoparticles over coverages obtained by the O2-based chemistry. The method might only be viable for Pt and not for the preparation of Ru and Pd nanoparticles and there alloys (with Pt). It is also questionable whether the N2* plasma approach allows for the preparation of the nanoparticles on highly structured materials (plasma cannot penetrate such materials). Moreover, it is not clear how the results rely on the specific reactor conditions employed by the authors. I can imagine that the N2* plasma conditions are very system-dependent and it is not obvious that similar results can be obtained by others. It might also only work SiO2 supports and not on other support materials. Finally, another vital point, and perhaps the most important one, is that the authors have not demonstrated the catalytic activity of the nanoparticles at all. Considering the existing literature, I think that this should be a requirement for publication of the results in a high-impact journal.
To summarize, I don't question the novelty of the claim of this manuscript but I do question whether the impact of the claim is really demonstrated. The manuscript will only have sufficient impact and be of wide interest to the community at large if the improved catalytic performance of nanoparticles prepared by this two-step method is really demonstrated.
Finally, there is another point I would like to raise: to my opinion, the introduction seems to be biased. I don't think it is sufficiently comprehensive and it does not acknowledge the major achievements within the field.
Reviewer #3 (Remarks to the Author): The manuscript is written well. It proposes a strategy to allow independent control of Pt particle size and coverage for nano-sized supported particles using a combination of O2-based and N2 plasma-based atomic layer deposition (ALD). Using the ALD method to synthesize model Pt nanoparticles has a lot of advantages compared to the use of more classical methods. The ALD method is a precise deposition techniques with good control over e.g. conformality, thickness, and composition. The field To the reviewers best knowledge the authors report independent tuning of size and coverage of Pt nanoparticles for the the first time. This is not only of great interest in the field of heterogeneous catalysis but also e.g. in surface science and micro-electronics. It could be speculated that a similar strategy can be used to deposit other (combinations of) metals.
The manuscript shows the independent control of size and coverage using a variety of techniques (GISAXS, XRD, SEM, HAADF-STEM) appropriate to characterize Pt particles. The data presented in the manuscript supports the conclusions well.
Putting the details of the GISAXS simulations in the supplementary information is a good decision. It would be beneficial to corroborate some of the assumptions in the GISAXS analysis. E.g. the validity of the assumption of a log-normal distribution function (line 46 p.3, Suppl. Inf.) and a relative width of σ = 1.1. The GISAXS simulations show good qualitative agreement with the experimental data, however a few cross-sections through the 2-dimensional data for e.g. constant q_y or q_z could also show a quantitative comparison (for figures S2, S3, S4, S5, S6). In this way the validity of the implicit choice of fixing the size distribution and fitting the shape (in stead of the other way around) could be shown. As GISAXS is the main technique used to extract relevant parameters from the experiments, a quantitative comparison between simulations and experiments will strengthen the authors conclusions. As all 2 dimensional data is available it will not generate much extra work for the authors to show a quantitative comparison.
A similar quantitative analysis of the SEM results (in stead of the more limited qualitative analysis shown on page 15, line 256 and in figures 5b and 6b, will make the claims even more convincing. Therefore, I recommend publication with minor additions to the data analysis of GISAXS and SEM.
We agree with the reviewer that Schwartzkopf et al. showed convincing proof for the ability of GISAXS to deduce the contact angle of truncated spheroidal particles in a study of a gold sputtering process [Nanoscale 5, 5053, 2013]. Based on simulations they showed a clear relation between the spheroid contact angle and the position of the second order of the first height peak in the scattering pattern, allowing to extract the nanoparticle contact angle based on the position of this second order scattering feature. However, in the recorded patterns for the Pt ALD processes, no such second order scattering feature can be observed. This suggests that the contact angle of the particles differs from particle to particle, causing smoothening of this second order scattering feature.
As requested by the reviewer, we have added Supplementary Figure 6 that compares 2D patterns simulated for different spheroidal Pt nanoparticles. The best agreement with the experiment was found when a mixture of full spheroids and hemispheroids is used for form factor calculation. We believe that this mixture of two different wetting conditions in the simulations corresponds to a real situation where the contact angle of the spheroidal Pt nanoparticles with the SiO 2 surface varies from particle to particle.
To further justify our choice of particle shape, we performed an electron tomography characterization of ALD-grown Pt nanoparticles ( Figure 2 in the main text). This "3D TEM" imaging technique confirmed the spheroidal shape of the Pt nanoparticles, in agreement with GISAXS. Although the 3D tomogram of the Pt nanoparticles suggests that the contact angle of the Pt nanoparticles indeed varies from particle to particle, one should be careful with this kind of interpretation due to so-called missing wedge artifacts during the reconstruction of the TEM tomography.
Finally, we agree with the reviewer that the particle shape might change with Pt loading. We indeed noticed that the agreement between experiment and simulations is worse for Pt loadings that are higher than those reported in this work, i.e. when the Pt cluster morphology is dominated by wormlike and coalesced Pt structures, as revealed by SEM imaging. However, for all GISAXS patterns and Pt loadings considered in this work, the morphology is mostly dominated by isolated nanoparticles and in this case, the assumed two-particle model yields reasonable agreement with the experimental data.
Changes to the manuscript: The following paragraph has been added to the Supplementary Information: "Finally, to motivate our two-particle model for calculating the form factor, Supplementary Figure 6 compares 2D patterns simulated for different spheroidal Pt nanoparticle geometries. In these simulations, the values for D, H and W, and those for ω and σ R are kept constant, but the form factor is calculated for 100% full spheroids, 100% hemispheroids or a 50 to 50% mixture of both particle types. In the experimental 2D GISAXS pattern, one observes next to two clear scattering peaks a less intense arc-like feature (marked by 1 in Supplementary Figure 6) and a triangular scattering that emerges from the main scattering peak (marked by 2). Note that these scattering features are apparent in most of the experimental patterns recorded in this study. However, in the simulated scattering patterns for the one-particle models, one observes only one of these scattering features. In case of 100% full spheroids, the pattern is marked by a clear arc-like feature. In case of 100% hemispheroids, a clear scattering feature emerges from the main peak. By assuming a mixture of the two particle types for calculation of the form factor, the appearance of the two scattering features, as observed in the experimental patterns, can be reproduced in the simulations. We believe that this mixture of two different wetting conditions in the simulations corresponds to a real situation where the contact angle of the spheroidal Pt nanoparticles with the SiO 2 surface varies from particle to particle, as also suggested by the TEM tomography result." Supplementary Figure 6 | Effect of particle shape on GISAXS simulations. Comparison between experimental and simulated GISAXS patterns calculated for different spheroidal particle shapes. The particle shape assumed for calculation of the form factor is displayed in the top right corner of the respective simulated 2D GISAXS pattern. The table includes the input parameters that were used for the calculations.
The following paragraph has been added to the main text: "Analysis of the GISAXS data through comparison with simulations assumes a certain shape for the Pt nanoparticles. Therefore, to obtain more insights in the 3D shape of the nanoparticles, an electron tomography study is performed on a sample with a Pt loading of ~45 atoms / nm 2 prepared by the O 2 -based ALD process. While conventional transmission electron microscopy (TEM) only yields a 2D projection of a 3D object, electron tomography allows reconstructing the 3D structure of the object based on a large number of 2D projection images. 49 To acquire a full tilt range of 2D projection images with high angle annular dark field (HAADF) scanning transmission electron microscopy (STEM), a plan-view sample is prepared as explained in the Methods section and mounted on a dedicated tomography holder. After acquisition and alignment of the HAADF-STEM images, the "Simultaneous Iterative Reconstruction Technique" (SIRT) is used for the reconstruction of the 3D structure of the specimen. The reconstructed volume of Pt nanoparticles deposited in a ca. 60 by 60 nm 2 region on the Si/SiO 2 surface is visualized in Figure 2. An animated version of the tomogram is provided in the Supplementary Information as a video. The majority of the Pt clusters have a spheroidal shape, while some clusters consist of agglomerated smaller particles. Rounded particles are expected to expose many atomic steps and kinks presenting high catalytic activity. 50,51 In the animated version of the tomogram, it can also be observed that some of the particles exhibit a flat surface at the Pt/SiO 2 interface (particles indicated by an arrow in Figure 2), while others seem to be full spheroids. The tomography study thus suggests that the contact angle of the Pt nanoparticles with the surface varies from particle to particle. However, it should be kept in mind that the tomography series was carried out on a planview TEM sample as indicated previously. Due to missing wedge artifacts, 49 which are more pronounced at the interface between the nanoparticles and the substrate in the case of plan" The following sentences have been added to the main text: "As exemplified in the Supplementary Information, best agreement with the experimental GISAXS patterns is obtained when a two particle model is used to describe the spheroidal particles. A mixture of 50% (75%) full spheroids and 50% (25%) hemi-spheroids is assumed to simulate the patterns for the O 2 -(N 2 * -)based ALD process. This mixture of two different wetting conditions in the simulations suggests a real situation where the contact angle of the spheroidal Pt nanoparticles with the SiO 2 surface varies from particle to particle, as also suggested by the tomography result. Moreover, GISAXS indicates (on average) larger contact angles (larger dewetting) for the Pt nanoparticles deposited via the N 2 * -based ALD process." 5) Actually, the used mixed particle shapes of full and hemi spheroids make me wonder and are unexpected. Justification will be necessary beyond best agreement with the data, since such wetting conditions are very unexpected. Somehow I do not see this in the real space realizations in the main manuscript.

Author reply:
The idea to combine two different particle shapes in the simulations was based on a previous publication by Kaune et al. who used a model consisting of parallelepiped and spheroid particle geometries to describe the cluster shape of gold nanoparticles [ACS Appl. Mater. Interf., 1, 353, 2009]. As mentioned above, we believe that the mixture of two different wetting conditions in the simulations corresponds to a real situation where the contact angle of the spheroidal Pt nanoparticles with the SiO 2 surface varies from particle to particle. This has been confirmed by an electron tomography study, though one should keep in mind that the morphology of the Pt nanoparticles near the SiO 2 interface might vary slightly from what is obtained in tomography due to missing edge artifacts.
The reviewer indicates that the varying contact angle was not incorporated in the real space realizations of the GISAXS analyses. We agree that this might cause some confusion and have therefore updated the real space sketches.

Changes to the manuscript:
The real space sketches in the main text have been updated and include now also the variation in contact angle over the different nanoparticles.
6) Error bars should be shown (e.g. figures 2a and 2d). Also the used size distributions should be elucidated.

Author reply:
The reviewer advises to add error bars to the average morphological parameters that were extracted from GISAXS. However, determining the error on these values is not straightforward. Therefore, this comment by the reviewer has been implemented in the Supplementary Information by adding a figure that shows how sensitive GISAXS is to small variations of the average particle height and width. Figure 4 gives the reader a visual impression of how accurate the average particle dimensions can be determined with our GISAXS analysis method.

Supplementary
Secondly, the reviewer advised us to elucidate the used size distribution. We apologize that this was not well explained before. We have added an explanation and figure to the Supplementary Information to make this clearer to the reader. Note that the average morphological parameters that are mentioned in the main text are determined without assuming a particle size distribution, as is explained in detail in the Supplementary Information. The distribution in particle sizes is only introduced to generate the simulated 2D patterns, which are compared to the experimental data in order to validate our GISAXS analysis strategy.
In addition, as advised by Reviewer #3, we have added a quantitative analysis of the SEM images shown in Figure 6. Lognormal functions have been fitted to the derived particle size distributions (Supplementary Figure 12) as well as to the particle size distribution obtained from TEM in Figure  Changes to the manuscript: The following paragraph has been added to the Supplementary Information: "To demonstrate how sensitive GISAXS is to changes in the parameters H and W, Supplementary Figure 4 illustrates the effect of systematic Ångstrom-level changes on the simulated patterns. The simulations show that changes in the particle height of 2Å can easily be distinguished by their change in oscillation period in the vertical line profile ((a), right graph). Similarly, a 2Å deviation in the particle radius (4Å in particle width) is shown to have a noticeable effect on the horizontal line profile ((b), left graph). In both cases, also the corresponding horizontal/vertical line profile has changed, though to a lesser extent." Supplementary Figure 4 | Sensitivity of GISAXS to changes in the average particle sizes. The sensitivity of GISAXS to particle height (a) and particle width (b) variations: experimental (black data points) and calculated (green, red and blue curves) 1D horizontal (left graph, q z = 0.722 nm -1 ) and vertical (right graph, q y = 0.59 nm -1 ) line profiles. The table includes the input parameters that were used for the calculations. For form factor calculation, a mixture of 50% full spheroids and 50% hemispheroids was used.
The following sentences have been added to the Supplementary Information: "To improve the agreement between simulation and experiment, the model that was used in step 2 to calculate the 1D line profiles is extended to account for the distribution in particle sizes. For the sake of simplicity, the particle height and width distributions are chosen to be coupled, in the sense that a distribution of particle radii at constant height/radius ratio implies also a distribution of particle heights. A lognormal distribution is assumed for the particle radius R, based on precedence in the literature 7-10 : with σ R the dimensionless geometric standard deviation. The size distribution is kept equal for both types of particles in the model (full spheroids and hemispheroids). The calculations furthermore use the local monodisperse approximation (LMA) formalism, which is commonly used for polydispersed systems. 5 As an illustration, Supplementary Figure 5 compares simulations with and without size distribution for the same sample as in Supplementary Figures 3 and 4. The obvious effect of the size distribution is smoothening of the 1D line profiles, leading to an improved agreement with the experimental data for a σ R -value of 1.1. Since the aim of the complete 2D simulations is to validate the derived values for D, H and W rather than to derive the exact width of the particle size distribution, the σ R -parameter was not treated as a fitting parameter but is kept constant to 1.1 for all simulations performed in this study."

Supplementary Figure 5 | Effect of size dispersion on GISAXS simulations. Comparison between experimental and simulated GISAXS patterns calculated without and with coupled size distribution for the particle width and height: (left) 2D GISAXS patterns, (right) experimental (black data points) and
calculated (red curves) 1D horizontal (top graph, q z = 0.722 nm -1 ) and vertical (bottom graph, q y = 0.59 nm -1 ) line profiles. The particle radius distribution is displayed in the top right corner of the respective simulated 2D GISAXS pattern. The other input parameters for the calculations are the same as those for the calculations in Supplementary Figure 4. For form factor calculation, a mixture of 50% full spheroids and 50% hemispheroids was used.
The following paragraph has been added to the Supplementary Information: "As shown in Figure 8 in the main text, a good agreement is found between the average particle radius obtained from TEM analysis and the one derived from the GISAXS analysis. Supplementary Figure 12 below presents additional analysis results for the SEM images included in Figure 6 of the main text, confirming again the agreement in average particle radius obtained from real-space electron microscopy measurements and reciprocal space GISAXS data. The black lines for samples A, B and C are fitted lognormal functions to the particle size distributions. The wide distribution observed for sample D is a consequence of the formation of wormlike structures when a large number of O 2 -based ALD cycles is applied. For all lognormal fits, the value for the dimensionless geometric standard deviation σ R is ~1.30. Similar fits to the size distributions obtained from TEM ( Figure 8) yield a σ R -value of ~1.25. Both of these values are larger than the value of 1.1 evaluated from GISAXS. However, for GISAXS simulations with a σ R -value of 1.25 or 1.30, the scattering features are highly smoothed or damped, in disagreement with the experimental patterns. Similar differences in particle radius distribution obtained from TEM and GISAXS have been observed before for 1-10 nm Au nanoparticles embedded in a SiO 2 film and may be attributed to different sampling conditions. 12 For our SEM and TEM analyses, 300 to 1000 particles are measured from a small region of the sample (< 500 x 500 nm 2 ) while GISAXS probes a sample area of ca. 300 nm x 2 cm, averaging over an estimated 10 8 particles." 7) The authors use a 2D paracrystal model, which is very unfortunate because it is not existing as proven by Wilhem Ruland in Makromol. Chem. 177, 3601-3617 (1976). Only a 1D paracrystal model is meaningful. Will this make any problem in the simulations?

Author reply and changes to the manuscript:
We thank the reviewer for this very valuable comment. In fact, all simulations were done using a 1D paracrystal model in IsGISAXS. We mistakenly mentioned 2D paracrystal model because we assume a regular 2D lattice in the first step of the GISAXS analysis. However, for all calculations in IsGISAXS, we have used the 1D paracrystal model, as is now mentioned correctly in the Supplementary Information.
8) The differences between data and simulation in the low q y region is NOT due to a beamstop problem but due to the used modeling. It means that in the model large scattering objects are existing which cannot be found in the real samples. They might be caused from the tail in the size distribution function and likely can be eliminated by truncating them. I do not consider this as a very big problem, but the authors should at minimum give a proper explanation instead of the beamstop story.

Author reply:
We thank the reviewer for this comment. The comment on the beam stop effect was actually based on experimental observations where a slightly different position of our in vacuum beam stop did have some effect on the horizontal cut in the low q y region. However, we agree that this might not be the main explanation for the discrepancy near q y = 0 nm -1 , and that the explanation offered by the reviewer is more appropriate.

Changes to the manuscript:
The following sentence has been removed from the Supplementary Information: "The discrepancy with the experimental images near q y = 0 nm -1 could result from beam stop shadowing." The following sentence has been added to the Supplementary Information: "The discrepancy with the experimental images near q y = 0 nm -1 arises from the interference function in the simulations showing a tail towards low q y -values originating from larger, more widely spaced particles which are not present in the real samples."

Reviewer #2 (Remarks to the Author):
The manuscript reports on that what is exactly expressed by the title: The independent tuning of size and coverage of supported Pt nanoparticles using ALD. In principle this is an interesting feature that builds on extensive other work that has been published in this area: -The preparation of supported nanoparticles by ALD for catalysis applications -The precise size-control of these nanoparticles (also for particles smaller than reported here) -The preparation of nanoparticles consisting of several materials (Pt, Ru, Pd), also in mixed phases (alloys) or in core/shell configurations -The preparation of nanoparticles on highly structured materials (e.g. on nanosphere supports) -The protection of the nanoparticles by overcoatings -The area-selective ALD of nanoparticles -The demonstration of the activity of ALD-prepared nanoparticles in several heterogenous catalysis reactions (viz. various dehydrogenation and oxidation reactions) -Etc.
This work goes back to the basis of the field and the (only) novelty of the work is that the authors show that the use of a N 2 * plasma as reactant allows for increasing the size of the Pt nanoparticles without changing the coverage of the Pt nanoparticles in terms of the number of particles per surface area. With the common chemistry using O 2 as the reactant, the size can be controlled but not without affecting the coverage of the nanoparticles.
As mentioned, this is a nice feature but by itself it does not warrant the publication of the work in Nature Communications. More important achievements (see above) have already been reported in high impact journals (including several Nature and Science journals). Furthermore, it is not clear what the impact of this work exactly is.

Author reply:
We thank the reviewer for the careful evaluation of our work. We regret that the reviewer is not convinced by the significance and novelty of our work.
We agree with the reviewer that 'ALD for catalysis' is a growing research field and that several important works have been published before: -the preparation and size-control of supported nanoparticles, e.g. citations 25, 28-33 -the preparation of alloyed and core/shell nanoparticles, e.g. citations 20-24 -the preparation of nanoparticles on high surface area supports, e.g. citations 28-33 -the protection of the nanoparticles by overcoatings, e.g. citations 26, 27 -the area-selective ALD of nanoparticles, e.g. citation 24 -the demonstration of the activity of ALD-prepared nanoparticles in several heterogenous catalysis reactions, e.g. citations 28-33 To stress the novelty of our work, we would like to respectfully remark that this work presents -an accurate tuning strategy to independently control the Pt nanoparticle size and coverage, even at high surface densities of nanoparticles for which precise control is often difficult to achieve due to easy merging and sintering of the Pt nanoparticles; -the first application of the N 2 * -based Pt ALD process for the growth of Pt nanoparticles; -the first in situ characterization of the evolution in morphology during ALD of Pt and of noble metals in general, yielding insights in Pt particle ALD growth with a level of detail missing so far; -the first convincing experimental proof to date of the important role of atom and cluster surface diffusion during the commonly applied O 2 -based Pt ALD process; -clear experimental evidence of the important role of the choice of reactant used in noble metal ALD.
The reviewer mentions that it is not clear what the impact of the work exactly is. We are convinced that the manuscript presents novel insights in nanoparticle growth by ALD that are highly important to the development of model systems for catalysis research (the first two points listed above). The tuning strategy that is presented in this work will be useful to create systems that allow to elucidate the effect of particle proximity on (electro)catalytic activity and selectivity. This has proven challenging by using conventional synthesis methods, such as incipient wetness or precipitation, because the effect of the particle size (distribution) cannot be scrutinized independently from the nanoparticle coverage, and in turn, the particle distance. To illustrate that the particle distance is indeed a determining parameter which receives attention in current catalysis research, we refer to the works by Nesselberger et al. . The former study revealed that the edge-to-edge particle distance between Pt clusters decisively influences the electrochemical oxygen reduction reaction (ORR) activity, while the latter study demonstrated that the particle distance between Cu nanoparticles plays a defining role in product selectivity during electrocatalytic reduction of CO 2 . We are convinced that the proposed ALD method offers an important novel strategy to further deconvolute the effect of nanoparticle size and distance in (electro)catalytic reactions.
Moreover, controlling the growth of noble metal ALD processes is not only important to the field of catalysis, but also to the field of microelectronics, as also indicated by referee 3. This manuscript presents an in-depth in situ characterization of the nucleation-controlled growth mode of Pt ALD and provides novel fundamental insights (the latter two point listed above) which broaden our understanding of noble metal ALD, important for catalysis and microelectronics, and which will inspire other researchers who focus on experimental or modeling studies of the nucleation of noble metal ALD processes.
The size of the Pt nanoparticles synthesized is relatively large (for optimized catalytic reactivity) whereas the 2-step method does not allow to increase the coverage of the nanoparticles over coverages obtained by the O2-based chemistry.

Author reply:
We respectfully remark that the highest coverage is obtained with the N 2 * -based ALD process, and not by the O 2 -based process as mentioned by the reviewer. Nevertheless, the reviewer is right that the introduced two-step method does not lead to higher coverages. However, the claim of the manuscript is not that the two-step process yields the most optimal morphologies for catalysis, but that the tuning potential in itself provides opportunities for fundamental catalysis research, as motivated above.
The reviewer mentions that the size of the Pt nanoparticles is relatively large. The lower limit for the particle sizes shown in this work, ca. 3-4 nm in diameter, is determined by the limit for which analysis of the GISAXS patterns can be performed in a reliable way. For lower Pt loadings, and thus smaller Pt nanoparticle sizes, a broad scattering peak without side minima/maxima is observed in GISAXS which is difficult to analyze. However, qualitative comparison of the scattering data at low Pt loadings for the O 2 -based vs. N 2 * -based Pt ALD process reveals similar trends as reported in the manuscript for Pt loadings > 45 atoms / nm 2 . The figure below shows horizontal line profiles through the scattering data and it can be observed that the peak maximum appears at higher q y -values for the N 2 * -based process (dashed lines, q y,max  1.0 nm -1 ) than for the O 2 -based process (solid lines, q y,max  0.75 nm -1 ). This qualitative comparison suggests that the N 2 * -based process results in smaller nanoparticles with a smaller particle-to-particle distance than the O 2 -based process, also at these low Pt loadings. The method might only be viable for Pt and not for the preparation of Ru and Pd nanoparticles and there alloys (with Pt).

Author reply:
As indicated by the reviewer, the presented tuning method will not be directly applicable to other noble metal ALD processes due to different chemistries. However, the manuscript does contain results that are important and relevant to the whole field of noble metal ALD. There are several noble metal ALD processes -for Ru, Os, Rh, and Ir -that use O 2 as a reactant at deposition temperatures above 200 ˚C. Researchers so far have not paid much attention to the impact of O 2 on the surface mobility, coalescence regime (dynamic vs. static) and evolution in nanoparticle coverage. However, as shown here for the Pt ALD process, O 2 might have a strong influence on the particle growth and one should be aware that tuning the size by changing the number of ALD cycles might come at the cost of a decrease in particle coverage. Therefore, the results presented here will trigger new investigations on the role of O 2 in other noble metal ALD processes.
In addition, we see opportunities that similar tuning strategies, based on changing the reactant, can be developed for other noble metal ALD processes. For example, for the deposition of Pd nanoparticles, the Pd(hfac) 2 precursor can be combined with formalin, H 2 , H 2 plasma or H 2 plasma followed by O 2 plasma. Since the O 2 plasma step in this latter process might cause diffusion-mediated coalescence like observed here for the oxidative Pt ALD chemistry, this could offer opportunities for tuning particle size and coverage by combining processes using different reactants.
Changes to the manuscript: The following sentences have been added to the Discussion section of the manuscript: "A final concluding remark is that the insights presented here for the particle growth during the O 2 -based ALD process for Pt may be equally relevant to other noble metal ALD processes using O 2 as a reactant for temperatures above 200˚C (e.g. for Ru, Os, Rh, Ir). 60 This work might therefore motivate future experimental studies exploring the influence of O 2 on the surface atom and cluster mobility during noble metal ALD." It is also questionable whether the N 2 * plasma approach allows for the preparation of the nanoparticles on highly structured materials (plasma cannot penetrate such materials). Moreover, it is not clear how the results rely on the specific reactor conditions employed by the authors. I can imagine that the N 2 * plasma conditions are very system-dependent and it is not obvious that similar results can be obtained by others.

Author reply:
We agree with the reviewer that coating mesoporous materials presents a higher challenge for plasma-enhanced compared to thermal ALD. On the other hand, we would like to emphasize that PE-ALD can be applied successfully on high surface area ( It might also only work SiO 2 supports and not on other support materials.

Author reply: The reviewer is right that the interaction between the Pt nanoparticles and the underlying surface differs from support to support, influencing the particle nucleation and growth during ALD. However, this comment is valid for all ALD-based strategies related to nanoparticle growth. On the other hand, using our in situ methodology, we have monitored Pt ALD growth on different metal oxide surfaces and, although there are support-related effects, these results prove that the O 2 -based ALD process is governed by a diffusion-mediated growth, marked by a decrease in number of particles per surface area with increasing number of ALD cycles, on all tested supports (SiO 2 , TiO 2 , Al 2 O 3 ). The figures below
show the scattering data measured in situ during growth on TiO 2 and Al 2 O 3 surfaces. The shift of the main scattering peak to lower q y -values with increasing number of ALD cycles is indicative of an increasing center-to-center distance, and in turn, decreasing particle coverage. These results give a strong indication that the presented tuning strategy is extendable to other metal oxide surfaces.

Figure: Selection of experimental 2D GISAXS images measured in situ during O 2 -based Pt ALD on Al 2 O 3 . The number in the top right corner is the ALD cycle number.
Finally, another vital point, and perhaps the most important one, is that the authors have not demonstrated the catalytic activity of the nanoparticles at all. Considering the existing literature, I think that this should be a requirement for publication of the results in a high-impact journal.

Author reply:
As suggested by the reviewer, the catalytic activity of the Pt nanoparticles has been evaluated in a proof-of-principle experiment. Because nanoparticle sintering is an often encountered problem during catalytic reactions at elevated temperatures (which may be prevented with an oxide ALD overcoat) and the focus of this work is on the differences in as-deposited morphology with the choice of reactant in the ALD process, the activity of the Pt nanoparticles was tested in electrocatalysis. The electrochemical hydrogen evolution reaction (HER) was selected as probe reaction.
The electrocatalytic experiment required the deposition of Pt nanoparticles on a conductive support. As also discussed above, we have clear indications that the Pt nanoparticle growth behavior, as reported in the manuscript for a SiO 2 support, is similar on other metal oxide supports. Therefore, fluorine-doped tin oxide (FTO) coated glass was selected as a conductive oxide support. Low Ptloading catalysts were prepared by depositing 3.5 μg Pt per cm 2 of FTO/glass substrate using both the O 2 -based and N 2 * -based ALD processes. Both ALD processes resulted in Pt nanoparticles that are effective catalysts for water splitting, with a better performance for the N 2 * -based sample.

Changes to the manuscript:
The following paragraphs have been added to the main text: "To demonstrate that the choice of reactant in Pt ALD is a determining parameter not only for the morphology of the deposited nanoparticles but also for their catalytic activity, Pt nanoparticles deposited via the O 2 -based and N 2 * based Pt ALD processes are evaluated in the hydrogen evolution reaction (HER), which is one half reaction in water electrolysis. Platinum is known to be the most active catalyst for the HER in acidic media. 4,5 Dispersing the Pt into nanoparticles, so as to maximize the surface over volume ratio, is a common way to enhance activity and to reduce the Pt content and cost. 4,5,57-59 Fluorine-doped tin oxide (FTO) coated glass is selected as a conductive oxide support for the fabrication of Pt ALD electrodes. Electrodes with low Pt loading (3.5 µg/cm 2 ) are prepared by tuning the number of ALD cycles of both the O 2 -based and N 2 * -based process. The loading corresponds to ca. 107 Pt atoms / nm 2 of glass substrate; note, the Pt loading per nm 2 of FTO surface will be lower due to surface roughness. (Figure 5(d)), suggesting a similar diffusion-mediated particle growth mechanism for the O 2 -based process on FTO. Figure 14). In the scanned potential region, the bare FTO coated glass is inert. Both the O 2 -based and N 2 * -based Pt ALD electrodes show HER activity, with a slightly better performance for the N 2 * -based sample. At an overpotential of 50 mV, the N 2 * -based sample and the O 2 -based sample exhibit a current density of 13.3 mA/cm 2 and 11.5 mA/cm 2 , respectively. The mass activities are comparable to other ALD-prepared Pt electrocatalysts, despite higher mass loadings. 4 Our results show that mastering Pt particle size by ALD offers the potential for fine-tuning state-of-the-art HER catalysts. The choice of reactant in the Pt ALD process influences the electrochemical activity, most likely due to a difference in nanoparticle morphology, i.e. nanoparticle shape, size and coverage."

The electrochemical characterizations are performed in 0.5 M H 2 SO 4 using a three-electrode cell. The activity of the catalysts is determined via cyclic voltammetry at a scan rate of 2 mV/s (Supplementary
The To summarize, I don't question the novelty of the claim of this manuscript but I do question whether the impact of the claim is really demonstrated. The manuscript will only have sufficient impact and be of wide interest to the community at large if the improved catalytic performance of nanoparticles prepared by this two-step method is really demonstrated.

Author reply:
As discussed above, the catalytic activity of the Pt nanoparticles synthesized with the O 2 -based and N 2 * -based ALD processes is now demonstrated in the manuscript. However, we would like to clarify our opinion on the expected impact of the manuscript. What we demonstrate in the manuscript is that we can accurately control the Pt nanoparticle size and coverage by using different reactants in the Pt ALD process and applying the correct sequences. We do not claim in the manuscript that the two-step method will (necessarily) lead to improved catalytic performance. Instead, we envision that the tuning method will impact the field of fundamental catalysis research by offering a new way to synthesize precise model systems that allow to link catalytic (or catalysis-related) and morphological properties of the nanoparticles.
To support this statement, we show below a result of a recent study of some of the authors focusing on the coarsening behavior of Pt nanoparticles at elevated temperatures. The coarsening of nanoparticles is considered the main cause for thermal deactivation and lifetime reduction of supported catalysts. The tuning method proposed in this manuscript was used to synthesize samples with a well-controlled equal amount of Pt atoms per surface area but distinct as-deposited morphology. These samples were annealed in 18% O 2 in He while in situ GISAXS patterns were recorded. The figure below shows the evolution in nanoparticle radius with sample temperature and reveals that the onset temperature for particle coarsening increases with increasing particle size. This example illustrates that the tuning strategy presented in this manuscript can indeed be used to synthesize model systems enabling systematic characterizations of the effect of particle morphology on catalysis-related processes. We are convinced that many other examples will follow. Finally, there is another point I would like to raise: to my opinion, the introduction seems to be biased. I don't think it is sufficiently comprehensive and it does not acknowledge the major achievements within the field.

Author reply:
As suggested by the referee, the introduction of the paper has been extended with a paragraph discussing prior work in the field of noble metal ALD for catalysis.
Changes to the manuscript: The following sentences have been added to the introduction: "This has led to several ALD-based strategies for the synthesis of monometallic and bimetallic nanoparticles. 14,

Reviewer #3 (Remarks to the Author):
The manuscript is written well. It proposes a strategy to allow independent control of Pt particle size and coverage for nano-sized supported particles using a combination of O 2 -based and N 2 plasma-based atomic layer deposition (ALD). Using the ALD method to synthesize model Pt nanoparticles has a lot of advantages compared to the use of more classical methods. The ALD method is a precise deposition techniques with good control over e.g. conformality, thickness, and composition.
To the reviewers best knowledge the authors report independent tuning of size and coverage of Pt nanoparticles for the the first time. This is not only of great interest in the field of heterogeneous catalysis but also e.g. in surface science and micro-electronics. It could be speculated that a similar strategy can be used to deposit other (combinations of) metals.
The manuscript shows the independent control of size and coverage using a variety of techniques (GISAXS, XRD, SEM, HAADF-STEM) appropriate to characterize Pt particles.
The data presented in the manuscript supports the conclusions well.

Author reply:
We thank the reviewer for the careful evaluation of our work and for the useful suggestions to further improve the manuscript. The manuscript has been modified according to the specific comments of the reviewer, as explained below.
Putting the details of the GISAXS simulations in the supplementary information is a good decision. It would be beneficial to corroborate some of the assumptions in the GISAXS analysis. E.g. the validity of the assumption of a log-normal distribution function (line 46 p.3, Suppl. Inf.) and a relative width of σ = 1.1. The GISAXS simulations show good qualitative agreement with the experimental data; however a few cross-sections through the 2-dimensional data for e.g. constant q y or q z could also show a quantitative comparison (for figures S2, S3, S4, S5, S6). In this way the validity of the implicit choice of fixing the size distribution and fitting the shape (instead of the other way around) could be shown. As GISAXS is the main technique used to extract relevant parameters from the experiments, a quantitative comparison between simulations and experiments will strengthen the authors' conclusions. As all 2 dimensional data is available it will not generate much extra work for the authors to show a quantitative comparison.

Author reply:
As suggested by the reviewer, we have added 1D horizontal and vertical line profiles to the Supplementary Figures 7 to 11 (before S2 to S6). These graphs show that the q y and q z -positions of the main scattering peak are well reproduced and also the positions of the side minima and maxima in the experimental data and in the simulations are in good agreement, indicating that the average morphological parameters used for the calculations are correct.
The assumption of a lognormal distribution function for the particle size distribution is based on precedence in the literature, as is now explicitly mentioned in the Supplementary Information. Note that the average morphological parameters that are mentioned in the main text are determined without assuming a particle size distribution, as is now better explained in the Supplementary  Information (e.g. Supplementary Figure 2). The distribution in particle sizes is only introduced to generate the simulated 2D patterns, which are compared to the experimental data in order to validate our GISAXS analysis strategy.
Changes to the manuscript: Supplementary Figures 7 to 11 have been updated with additional graphs showing 1D line profiles though the GISAXS experimental data and simulations.
The following sentences have been added to the Supplementary Information: "To improve the agreement between simulation and experiment, the model that was used in step 2 to calculate the 1D line profiles is extended to account for the distribution in particle sizes. For the sake of simplicity, the particle height and width distributions are chosen to be coupled, in the sense that a distribution of particle radii at constant height/radius ratio implies also a distribution of particle heights. A lognormal distribution is assumed for the particle radius R, based on precedence in the literature 7-10 : with σ R the dimensionless geometric standard deviation. The size distribution is kept equal for both types of particles in the model (full spheroids and hemispheroids)." The A similar quantitative analysis of the SEM results (instead of the more limited qualitative analysis shown on page 15, line 256 and in figures 5b and 6b, will make the claims even more convincing.

Author reply:
As suggested by the reviewer, the SEM images in Figures 6b and 7b (before 5b and 6b) have been analyzed in a more quantitative way with respect to the particle size distribution and particle coverage.
Changes to the manuscript: The following sentences have been added to the main text: "Indeed, quantitative analysis of the SEM images for samples b, c and d by manual counting the Pt nanoparticles in a 150 by 150 nm 2 area yields particle coverages of 1.58 . 10 12 cm -2 , 1.60 . 10 12 cm -2 and 1.53 . 10 12 cm -2 , respectively. These values correspond to an average center-to-center distance D of 8.0  0.1 nm. This value is higher than the one obtained from GISAXS analysis, 7.1 nm, which is likely due to the fact that agglomerated particles and particles at the edges of the SEM images are excluded from the particle count. For sample a, the contrast between the background and the small nanoparticles in SEM is insufficient to allow for a reliable particle count." The following paragraph has been added to the Supplementary Information: "As shown in Figure 8 in the main text, a good agreement is found between the average particle radius obtained from TEM analysis and the one derived from the GISAXS analysis. Supplementary Figure 12 below presents additional analysis results for the SEM images included in Figure 6 of the main text, confirming again the agreement in average particle radius obtained from real-space electron microscopy measurements and reciprocal space GISAXS data. 5) The authors write that the idea to combine two different particle shapes in the simulations was based on a previous publication by Kaune et al. who used a model consisting of parallelepiped and spheroid particle geometries to describe the cluster shape of gold nanoparticles [ACS Appl. Mater. Interf., 1, 353, 2009]. As a consequence, I think that is necessary to mention such existing approach from literature in the main manuscript and to explain the ideas behind. Other readers will have the same problem in understanding the meaning of such approach with mixed wetting conditions. It is good scientific practice to cite all used references.
6) The reviewer agrees that calculation of error bars is a difficult task. However, error bars are very important in judging the quality of extracted parameters. I cannot accept the absence of error bars and the added simulations showing sensitivity give a good first hint but cannot replace error bars.
Reviewer #3 (Remarks to the Author): The authors have given appropriate answers to the questions and remarks in my review and they have updated both the manuscript and the supplementary material, including figures. Also replies and changes to the manuscript based on input from the other referees has improved the manuscript. To my opinion the manuscript can be published.

Reviewer #4 (Remarks to the Author):
This is an excellent study describing a method to control the size and spacing of Pt nanoparticles on a SiO2 surface by ALD. The method combines two processes for ALD Pt: the first uses O2 as the co-reactant, and produces a nanoparticle spacing that increases with increasing ALD Pt cycles, and the second uses N2* as the co-reactant, and produces a constant nanoparticle spacing. The O2 process is used first to adjust the particle spacing and the N2* process is used next to tune the Pt nanoparticle size. The authors primarily use in situ GISAXS to demonstrate this control, but they further support their findings using SEM and STEM.
I have no arguments regarding the science in the paper, and I find that the authors have done a thorough job of replying to the technical comments made by the three reviewers. My only criticism of this paper relates to novelty. In the initial review, Reviewer 2 expressed a similar concern as follows: "…the (only) novelty of the work is that the authors show that the use of a N2* plasma as reactant allows for increasing the size of the Pt nanoparticles without changing the coverage of the Pt nanoparticles… this is a nice feature but by itself it does not warrant the publication of the work in Nature Communications…". For the most part, I agree with this statement. The effect of the N2* plasma is the most significant finding, but there are some additional "firsts" in this paper that the authors point out below in their rebuttal to this comment. However, as I describe below, I do not feel that the N2* process and the other accomplishments listed by the author meet the novelty criterion for publication in Nature Communications given that there are many previous publications that describe very similar work.
In response to this comment by Reviewer 2, and in defense of the novelty of their work, the authors present the following rebuttal (I have numbered the points): "To stress the novelty of our work, we would like to respectfully remark that this work presents (1) an accurate tuning strategy to independently control the Pt nanoparticle size and coverage, even at high surface densities of nanoparticles for which precise control is often difficult to achieve due to easy merging and sintering of the Pt nanoparticles; (2) the first application of the N2*-based Pt ALD process for the growth of Pt nanoparticles; (3) the first in situ characterization of the evolution in morphology during ALD of Pt and of noble metals in general, yielding insights in Pt particle ALD growth with a level of detail missing so far; (4) the first convincing experimental proof to date of the important role of atom and cluster surface diffusion during the commonly applied O2-based Pt ALD process; (5) clear experimental evidence of the important role of the choice of reactant used in noble metal ALD." There are a number of previous works, some of which were not referenced in the manuscript, that demonstrate to some degree each of these points as follows: (1) Independent control over particle size and coverage has been demonstrated previously (refs. 21, 23). In these studies, control was achieved by decoupling ALD on the substrate from ALD on existing metal particles.
(2) The authors have used N2* previously for Pt ALD films (ref. 37), and the extension to Pt nanoparticles is not surprising since Pt ALD nearly always produces nanoparticles during the nucleation stage. Given these previous papers, I do not feel that the manuscript is sufficiently novel for publication in Nature Communications.

Reviewer #1 (Remarks to the Author):
The authors have addressed most of my concerns in a convincing way. However, not all points are fully finished. The still open questions or concerns are numbered based on the numbering from the first report: Author reply: We thank the reviewer for the careful evaluation of our work and his/her appreciation regarding most of our previous revisions and responses. We have addressed the two remaining comments of the reviewer in the second revised version of the manuscript as explained below.
5) The authors write that the idea to combine two different particle shapes in the simulations was based on a previous publication by Kaune et al. who used a model consisting of parallelepiped and spheroid particle geometries to describe the cluster shape of gold nanoparticles [ACS Appl. Mater. Interf., 1, 353, 2009]. As a consequence, I think that is necessary to mention such existing approach from literature in the main manuscript and to explain the ideas behind. Other readers will have the same problem in understanding the meaning of such approach with mixed wetting conditions. It is good scientific practice to cite all used references.

Author reply:
We agree with the reviewer that we should have added a reference to the work of Kaune et al.

Changes to the manuscript:
A reference to Kaune et al. has been added to the manuscript: "As exemplified in the Supplementary Information, best agreement with the experimental GISAXS patterns is obtained when a two particle model is used to describe the spheroidal particles. 53 The following sentence and reference have been added to the Supplementary Information to explain the origin of the idea behind the two-particle model used for the GISAXS simulations: "It should be noted that a similar simulation approach was used before by Kaune et al. who reported a two-particle model consisting of parallelepipeds and spheroids to reproduce both the intensity distribution of the side peaks and the interconnecting streaks observed in experimental GISAXS patterns recorded for gold cluster growth on poly(N-vinylcarbazole). 12 6) The reviewer agrees that calculation of error bars is a difficult task. However, error bars are very important in judging the quality of extracted parameters. I cannot accept the absence of error bars and the added simulations showing sensitivity give a good first hint but cannot replace error bars.

Author reply:
As requested by the reviewer, substantial effort was made to estimate the uncertainties of the extracted parameters (mean particle height H, particle width W and center-to-center distance D). The uncertainties were estimated based on a quantitative comparison of experimental and calculated 1D vertical (for H) and horizontal (for W and D) line profiles. The sum of squared residuals (SSR) was calculated for varying input values for one of the parameters (while the other parameters were kept constant). An increase in the SSR of ca. 50% (with respect to the SSR obtained when using the values extracted from our GISAXS analysis as input) was judged to give a good measure for the accuracy of our analysis approach. As such, by analyzing the SSR against a range of input values, the uncertainties of the extracted parameters were determined and added as error bars in Figure 3 Figures 12b and 12c, respectively). In both cases, SSR values were calculated based on the comparison of experimental and simulated 1D horizontal line profiles. In these profiles, the value of D mainly influences the q y -position of the main scattering peak (via the interference function), while the q y -positions of the minima and second maximum mainly originate from the form factor and thus the value of W. Therefore, calculation of the SSR values for varying values of D was done by limiting the q y -range of the 1D horizontal line profiles to the main scattering peak (Supplementary Figure 12c). In contrast, calculation of the SSR values for varying values of W was done by excluding the q y -range of the main scattering peak from the 1D horizontal line profiles (Supplementary Figure  12b). For most of the W and D data points in Figure 3a, again a parabolic-type relation with a minimum near the extracted W and D values was found, allowing to estimate the uncertainty of these values by evaluating the simulations with a relative SSR of ca. 1.5, as explained above for the uncertainty of H and as illustrated in Supplementary Figures 12b and 12c, respectively. For the N 2 *based Pt ALD process and Pt loadings above ~160 atoms / nm 2 , it was not possible to obtain a full parabola-type curve when varying D because of the constraint that D > W. In these cases, the uncertainty was estimated by evaluating only one simulation with a relative SSR of ca. 1.5 and doubling the obtained offset in D.